1 /*
2 * NVMe admin command implementation.
3 * Copyright (c) 2015-2016 HGST, a Western Digital Company.
4 *
5 * This program is free software; you can redistribute it and/or modify it
6 * under the terms and conditions of the GNU General Public License,
7 * version 2, as published by the Free Software Foundation.
8 *
9 * This program is distributed in the hope it will be useful, but WITHOUT
10 * ANY WARRANTY; without even the implied warranty of MERCHANTABILITY or
11 * FITNESS FOR A PARTICULAR PURPOSE. See the GNU General Public License for
12 * more details.
13 */
14 #define pr_fmt(fmt) KBUILD_MODNAME ": " fmt
15 #include <linux/module.h>
16 #include <generated/utsrelease.h>
17 #include <asm/unaligned.h>
18 #include "nvmet.h"
19
nvmet_get_log_page_len(struct nvme_command * cmd)20 u32 nvmet_get_log_page_len(struct nvme_command *cmd)
21 {
22 u32 len = le16_to_cpu(cmd->get_log_page.numdu);
23
24 len <<= 16;
25 len += le16_to_cpu(cmd->get_log_page.numdl);
26 /* NUMD is a 0's based value */
27 len += 1;
28 len *= sizeof(u32);
29
30 return len;
31 }
32
nvmet_get_smart_log_nsid(struct nvmet_req * req,struct nvme_smart_log * slog)33 static u16 nvmet_get_smart_log_nsid(struct nvmet_req *req,
34 struct nvme_smart_log *slog)
35 {
36 u16 status;
37 struct nvmet_ns *ns;
38 u64 host_reads, host_writes, data_units_read, data_units_written;
39
40 status = NVME_SC_SUCCESS;
41 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->get_log_page.nsid);
42 if (!ns) {
43 status = NVME_SC_INVALID_NS;
44 pr_err("nvmet : Counld not find namespace id : %d\n",
45 le32_to_cpu(req->cmd->get_log_page.nsid));
46 goto out;
47 }
48
49 host_reads = part_stat_read(ns->bdev->bd_part, ios[READ]);
50 data_units_read = part_stat_read(ns->bdev->bd_part, sectors[READ]);
51 host_writes = part_stat_read(ns->bdev->bd_part, ios[WRITE]);
52 data_units_written = part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
53
54 put_unaligned_le64(host_reads, &slog->host_reads[0]);
55 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
56 put_unaligned_le64(host_writes, &slog->host_writes[0]);
57 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
58 nvmet_put_namespace(ns);
59 out:
60 return status;
61 }
62
nvmet_get_smart_log_all(struct nvmet_req * req,struct nvme_smart_log * slog)63 static u16 nvmet_get_smart_log_all(struct nvmet_req *req,
64 struct nvme_smart_log *slog)
65 {
66 u16 status;
67 u64 host_reads = 0, host_writes = 0;
68 u64 data_units_read = 0, data_units_written = 0;
69 struct nvmet_ns *ns;
70 struct nvmet_ctrl *ctrl;
71
72 status = NVME_SC_SUCCESS;
73 ctrl = req->sq->ctrl;
74
75 rcu_read_lock();
76 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
77 host_reads += part_stat_read(ns->bdev->bd_part, ios[READ]);
78 data_units_read +=
79 part_stat_read(ns->bdev->bd_part, sectors[READ]);
80 host_writes += part_stat_read(ns->bdev->bd_part, ios[WRITE]);
81 data_units_written +=
82 part_stat_read(ns->bdev->bd_part, sectors[WRITE]);
83
84 }
85 rcu_read_unlock();
86
87 put_unaligned_le64(host_reads, &slog->host_reads[0]);
88 put_unaligned_le64(data_units_read, &slog->data_units_read[0]);
89 put_unaligned_le64(host_writes, &slog->host_writes[0]);
90 put_unaligned_le64(data_units_written, &slog->data_units_written[0]);
91
92 return status;
93 }
94
nvmet_get_smart_log(struct nvmet_req * req,struct nvme_smart_log * slog)95 static u16 nvmet_get_smart_log(struct nvmet_req *req,
96 struct nvme_smart_log *slog)
97 {
98 u16 status;
99
100 WARN_ON(req == NULL || slog == NULL);
101 if (req->cmd->get_log_page.nsid == 0xFFFFFFFF)
102 status = nvmet_get_smart_log_all(req, slog);
103 else
104 status = nvmet_get_smart_log_nsid(req, slog);
105 return status;
106 }
107
nvmet_execute_get_log_page(struct nvmet_req * req)108 static void nvmet_execute_get_log_page(struct nvmet_req *req)
109 {
110 struct nvme_smart_log *smart_log;
111 size_t data_len = nvmet_get_log_page_len(req->cmd);
112 void *buf;
113 u16 status = 0;
114
115 buf = kzalloc(data_len, GFP_KERNEL);
116 if (!buf) {
117 status = NVME_SC_INTERNAL;
118 goto out;
119 }
120
121 switch (req->cmd->get_log_page.lid) {
122 case 0x01:
123 /*
124 * We currently never set the More bit in the status field,
125 * so all error log entries are invalid and can be zeroed out.
126 * This is called a minum viable implementation (TM) of this
127 * mandatory log page.
128 */
129 break;
130 case 0x02:
131 /*
132 * XXX: fill out actual smart log
133 *
134 * We might have a hard time coming up with useful values for
135 * many of the fields, and even when we have useful data
136 * available (e.g. units or commands read/written) those aren't
137 * persistent over power loss.
138 */
139 if (data_len != sizeof(*smart_log)) {
140 status = NVME_SC_INTERNAL;
141 goto err;
142 }
143 smart_log = buf;
144 status = nvmet_get_smart_log(req, smart_log);
145 if (status) {
146 memset(buf, '\0', data_len);
147 goto err;
148 }
149 break;
150 case 0x03:
151 /*
152 * We only support a single firmware slot which always is
153 * active, so we can zero out the whole firmware slot log and
154 * still claim to fully implement this mandatory log page.
155 */
156 break;
157 default:
158 BUG();
159 }
160
161 status = nvmet_copy_to_sgl(req, 0, buf, data_len);
162
163 err:
164 kfree(buf);
165 out:
166 nvmet_req_complete(req, status);
167 }
168
nvmet_execute_identify_ctrl(struct nvmet_req * req)169 static void nvmet_execute_identify_ctrl(struct nvmet_req *req)
170 {
171 struct nvmet_ctrl *ctrl = req->sq->ctrl;
172 struct nvme_id_ctrl *id;
173 u16 status = 0;
174
175 id = kzalloc(sizeof(*id), GFP_KERNEL);
176 if (!id) {
177 status = NVME_SC_INTERNAL;
178 goto out;
179 }
180
181 /* XXX: figure out how to assign real vendors IDs. */
182 id->vid = 0;
183 id->ssvid = 0;
184
185 memset(id->sn, ' ', sizeof(id->sn));
186 snprintf(id->sn, sizeof(id->sn), "%llx", ctrl->serial);
187
188 memset(id->mn, ' ', sizeof(id->mn));
189 strncpy((char *)id->mn, "Linux", sizeof(id->mn));
190
191 memset(id->fr, ' ', sizeof(id->fr));
192 strncpy((char *)id->fr, UTS_RELEASE, sizeof(id->fr));
193
194 id->rab = 6;
195
196 /*
197 * XXX: figure out how we can assign a IEEE OUI, but until then
198 * the safest is to leave it as zeroes.
199 */
200
201 /* we support multiple ports and multiples hosts: */
202 id->cmic = (1 << 0) | (1 << 1);
203
204 /* no limit on data transfer sizes for now */
205 id->mdts = 0;
206 id->cntlid = cpu_to_le16(ctrl->cntlid);
207 id->ver = cpu_to_le32(ctrl->subsys->ver);
208
209 /* XXX: figure out what to do about RTD3R/RTD3 */
210 id->oaes = cpu_to_le32(1 << 8);
211 id->ctratt = cpu_to_le32(1 << 0);
212
213 id->oacs = 0;
214
215 /*
216 * We don't really have a practical limit on the number of abort
217 * comands. But we don't do anything useful for abort either, so
218 * no point in allowing more abort commands than the spec requires.
219 */
220 id->acl = 3;
221
222 id->aerl = NVMET_ASYNC_EVENTS - 1;
223
224 /* first slot is read-only, only one slot supported */
225 id->frmw = (1 << 0) | (1 << 1);
226 id->lpa = (1 << 0) | (1 << 2);
227 id->elpe = NVMET_ERROR_LOG_SLOTS - 1;
228 id->npss = 0;
229
230 /* We support keep-alive timeout in granularity of seconds */
231 id->kas = cpu_to_le16(NVMET_KAS);
232
233 id->sqes = (0x6 << 4) | 0x6;
234 id->cqes = (0x4 << 4) | 0x4;
235
236 /* no enforcement soft-limit for maxcmd - pick arbitrary high value */
237 id->maxcmd = cpu_to_le16(NVMET_MAX_CMD);
238
239 id->nn = cpu_to_le32(ctrl->subsys->max_nsid);
240 id->oncs = cpu_to_le16(NVME_CTRL_ONCS_DSM);
241
242 /* XXX: don't report vwc if the underlying device is write through */
243 id->vwc = NVME_CTRL_VWC_PRESENT;
244
245 /*
246 * We can't support atomic writes bigger than a LBA without support
247 * from the backend device.
248 */
249 id->awun = 0;
250 id->awupf = 0;
251
252 id->sgls = cpu_to_le32(1 << 0); /* we always support SGLs */
253 if (ctrl->ops->has_keyed_sgls)
254 id->sgls |= cpu_to_le32(1 << 2);
255 if (ctrl->ops->sqe_inline_size)
256 id->sgls |= cpu_to_le32(1 << 20);
257
258 strcpy(id->subnqn, ctrl->subsys->subsysnqn);
259
260 /* Max command capsule size is sqe + single page of in-capsule data */
261 id->ioccsz = cpu_to_le32((sizeof(struct nvme_command) +
262 ctrl->ops->sqe_inline_size) / 16);
263 /* Max response capsule size is cqe */
264 id->iorcsz = cpu_to_le32(sizeof(struct nvme_completion) / 16);
265
266 id->msdbd = ctrl->ops->msdbd;
267
268 /*
269 * Meh, we don't really support any power state. Fake up the same
270 * values that qemu does.
271 */
272 id->psd[0].max_power = cpu_to_le16(0x9c4);
273 id->psd[0].entry_lat = cpu_to_le32(0x10);
274 id->psd[0].exit_lat = cpu_to_le32(0x4);
275
276 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
277
278 kfree(id);
279 out:
280 nvmet_req_complete(req, status);
281 }
282
nvmet_execute_identify_ns(struct nvmet_req * req)283 static void nvmet_execute_identify_ns(struct nvmet_req *req)
284 {
285 struct nvmet_ns *ns;
286 struct nvme_id_ns *id;
287 u16 status = 0;
288
289 ns = nvmet_find_namespace(req->sq->ctrl, req->cmd->identify.nsid);
290 if (!ns) {
291 status = NVME_SC_INVALID_NS | NVME_SC_DNR;
292 goto out;
293 }
294
295 id = kzalloc(sizeof(*id), GFP_KERNEL);
296 if (!id) {
297 status = NVME_SC_INTERNAL;
298 goto out_put_ns;
299 }
300
301 /*
302 * nuse = ncap = nsze isn't aways true, but we have no way to find
303 * that out from the underlying device.
304 */
305 id->ncap = id->nuse = id->nsze =
306 cpu_to_le64(ns->size >> ns->blksize_shift);
307
308 /*
309 * We just provide a single LBA format that matches what the
310 * underlying device reports.
311 */
312 id->nlbaf = 0;
313 id->flbas = 0;
314
315 /*
316 * Our namespace might always be shared. Not just with other
317 * controllers, but also with any other user of the block device.
318 */
319 id->nmic = (1 << 0);
320
321 memcpy(&id->nguid, &ns->nguid, sizeof(uuid_le));
322
323 id->lbaf[0].ds = ns->blksize_shift;
324
325 status = nvmet_copy_to_sgl(req, 0, id, sizeof(*id));
326
327 kfree(id);
328 out_put_ns:
329 nvmet_put_namespace(ns);
330 out:
331 nvmet_req_complete(req, status);
332 }
333
nvmet_execute_identify_nslist(struct nvmet_req * req)334 static void nvmet_execute_identify_nslist(struct nvmet_req *req)
335 {
336 static const int buf_size = 4096;
337 struct nvmet_ctrl *ctrl = req->sq->ctrl;
338 struct nvmet_ns *ns;
339 u32 min_nsid = le32_to_cpu(req->cmd->identify.nsid);
340 __le32 *list;
341 u16 status = 0;
342 int i = 0;
343
344 list = kzalloc(buf_size, GFP_KERNEL);
345 if (!list) {
346 status = NVME_SC_INTERNAL;
347 goto out;
348 }
349
350 rcu_read_lock();
351 list_for_each_entry_rcu(ns, &ctrl->subsys->namespaces, dev_link) {
352 if (ns->nsid <= min_nsid)
353 continue;
354 list[i++] = cpu_to_le32(ns->nsid);
355 if (i == buf_size / sizeof(__le32))
356 break;
357 }
358 rcu_read_unlock();
359
360 status = nvmet_copy_to_sgl(req, 0, list, buf_size);
361
362 kfree(list);
363 out:
364 nvmet_req_complete(req, status);
365 }
366
367 /*
368 * A "mimimum viable" abort implementation: the command is mandatory in the
369 * spec, but we are not required to do any useful work. We couldn't really
370 * do a useful abort, so don't bother even with waiting for the command
371 * to be exectuted and return immediately telling the command to abort
372 * wasn't found.
373 */
nvmet_execute_abort(struct nvmet_req * req)374 static void nvmet_execute_abort(struct nvmet_req *req)
375 {
376 nvmet_set_result(req, 1);
377 nvmet_req_complete(req, 0);
378 }
379
nvmet_execute_set_features(struct nvmet_req * req)380 static void nvmet_execute_set_features(struct nvmet_req *req)
381 {
382 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
383 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
384 u32 val32;
385 u16 status = 0;
386
387 switch (cdw10 & 0xf) {
388 case NVME_FEAT_NUM_QUEUES:
389 nvmet_set_result(req,
390 (subsys->max_qid - 1) | ((subsys->max_qid - 1) << 16));
391 break;
392 case NVME_FEAT_KATO:
393 val32 = le32_to_cpu(req->cmd->common.cdw10[1]);
394 req->sq->ctrl->kato = DIV_ROUND_UP(val32, 1000);
395 nvmet_set_result(req, req->sq->ctrl->kato);
396 break;
397 default:
398 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
399 break;
400 }
401
402 nvmet_req_complete(req, status);
403 }
404
nvmet_execute_get_features(struct nvmet_req * req)405 static void nvmet_execute_get_features(struct nvmet_req *req)
406 {
407 struct nvmet_subsys *subsys = req->sq->ctrl->subsys;
408 u32 cdw10 = le32_to_cpu(req->cmd->common.cdw10[0]);
409 u16 status = 0;
410
411 switch (cdw10 & 0xf) {
412 /*
413 * These features are mandatory in the spec, but we don't
414 * have a useful way to implement them. We'll eventually
415 * need to come up with some fake values for these.
416 */
417 #if 0
418 case NVME_FEAT_ARBITRATION:
419 break;
420 case NVME_FEAT_POWER_MGMT:
421 break;
422 case NVME_FEAT_TEMP_THRESH:
423 break;
424 case NVME_FEAT_ERR_RECOVERY:
425 break;
426 case NVME_FEAT_IRQ_COALESCE:
427 break;
428 case NVME_FEAT_IRQ_CONFIG:
429 break;
430 case NVME_FEAT_WRITE_ATOMIC:
431 break;
432 case NVME_FEAT_ASYNC_EVENT:
433 break;
434 #endif
435 case NVME_FEAT_VOLATILE_WC:
436 nvmet_set_result(req, 1);
437 break;
438 case NVME_FEAT_NUM_QUEUES:
439 nvmet_set_result(req,
440 (subsys->max_qid-1) | ((subsys->max_qid-1) << 16));
441 break;
442 case NVME_FEAT_KATO:
443 nvmet_set_result(req, req->sq->ctrl->kato * 1000);
444 break;
445 default:
446 status = NVME_SC_INVALID_FIELD | NVME_SC_DNR;
447 break;
448 }
449
450 nvmet_req_complete(req, status);
451 }
452
nvmet_execute_async_event(struct nvmet_req * req)453 static void nvmet_execute_async_event(struct nvmet_req *req)
454 {
455 struct nvmet_ctrl *ctrl = req->sq->ctrl;
456
457 mutex_lock(&ctrl->lock);
458 if (ctrl->nr_async_event_cmds >= NVMET_ASYNC_EVENTS) {
459 mutex_unlock(&ctrl->lock);
460 nvmet_req_complete(req, NVME_SC_ASYNC_LIMIT | NVME_SC_DNR);
461 return;
462 }
463 ctrl->async_event_cmds[ctrl->nr_async_event_cmds++] = req;
464 mutex_unlock(&ctrl->lock);
465
466 schedule_work(&ctrl->async_event_work);
467 }
468
nvmet_execute_keep_alive(struct nvmet_req * req)469 static void nvmet_execute_keep_alive(struct nvmet_req *req)
470 {
471 struct nvmet_ctrl *ctrl = req->sq->ctrl;
472
473 pr_debug("ctrl %d update keep-alive timer for %d secs\n",
474 ctrl->cntlid, ctrl->kato);
475
476 mod_delayed_work(system_wq, &ctrl->ka_work, ctrl->kato * HZ);
477 nvmet_req_complete(req, 0);
478 }
479
nvmet_parse_admin_cmd(struct nvmet_req * req)480 int nvmet_parse_admin_cmd(struct nvmet_req *req)
481 {
482 struct nvme_command *cmd = req->cmd;
483
484 req->ns = NULL;
485
486 if (unlikely(!(req->sq->ctrl->cc & NVME_CC_ENABLE))) {
487 pr_err("nvmet: got admin cmd %d while CC.EN == 0\n",
488 cmd->common.opcode);
489 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
490 }
491 if (unlikely(!(req->sq->ctrl->csts & NVME_CSTS_RDY))) {
492 pr_err("nvmet: got admin cmd %d while CSTS.RDY == 0\n",
493 cmd->common.opcode);
494 return NVME_SC_CMD_SEQ_ERROR | NVME_SC_DNR;
495 }
496
497 switch (cmd->common.opcode) {
498 case nvme_admin_get_log_page:
499 req->data_len = nvmet_get_log_page_len(cmd);
500
501 switch (cmd->get_log_page.lid) {
502 case 0x01:
503 case 0x02:
504 case 0x03:
505 req->execute = nvmet_execute_get_log_page;
506 return 0;
507 }
508 break;
509 case nvme_admin_identify:
510 req->data_len = 4096;
511 switch (le32_to_cpu(cmd->identify.cns)) {
512 case NVME_ID_CNS_NS:
513 req->execute = nvmet_execute_identify_ns;
514 return 0;
515 case NVME_ID_CNS_CTRL:
516 req->execute = nvmet_execute_identify_ctrl;
517 return 0;
518 case NVME_ID_CNS_NS_ACTIVE_LIST:
519 req->execute = nvmet_execute_identify_nslist;
520 return 0;
521 }
522 break;
523 case nvme_admin_abort_cmd:
524 req->execute = nvmet_execute_abort;
525 req->data_len = 0;
526 return 0;
527 case nvme_admin_set_features:
528 req->execute = nvmet_execute_set_features;
529 req->data_len = 0;
530 return 0;
531 case nvme_admin_get_features:
532 req->execute = nvmet_execute_get_features;
533 req->data_len = 0;
534 return 0;
535 case nvme_admin_async_event:
536 req->execute = nvmet_execute_async_event;
537 req->data_len = 0;
538 return 0;
539 case nvme_admin_keep_alive:
540 req->execute = nvmet_execute_keep_alive;
541 req->data_len = 0;
542 return 0;
543 }
544
545 pr_err("nvmet: unhandled cmd %d\n", cmd->common.opcode);
546 return NVME_SC_INVALID_OPCODE | NVME_SC_DNR;
547 }
548